Engineering scale bump aerodynamic noise prediction method based on data fusion

Abstract

The invention discloses an engineering scale bump aerodynamic noise prediction method based on data fusion, and belongs to the technical field of spacecraft aerodynamic design. The method comprises the following steps: firstly, acquiring wind tunnel test data, simulation data, corresponding test characteristic data and simulation characteristic data of each scaling model of a bulge; secondly, constructing a recurrent neural network, and forming a model for predicting test power spectral density data based on simulation power spectral density data by taking a mapping relation of active subspace vectors to simulation conditions as mapping from a neural network input layer to a first layer of an intermediate layer; and acquiring simulation data of the engineering scale bump, inputting the simulation data into the model, and predicting flight test noise power spectral density data of the engineering scale bump. Through the application of the method, the physical constraint and interpretability of the neural network are improved, the model confidence is improved, the aerodynamic noise of the projection in the engineering scale can be predicted more accurately, and the aircraft performance can be improved.

Description

Technical Field

[0001] The invention relates to a method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion, and belongs to the technical field of aerodynamic design of spacecraft. Background Art

[0002] In the field of aerospace, the prediction of aerodynamic noise of protrusions is of great significance to aircraft design and performance evaluation. However, due to the complexity of the problem and the limitation of computing resources, it is difficult for traditional numerical simulation methods to accurately analyze the turbulence phenomenon and the interaction process of the shock wave boundary layer upstream of engineering-scale protrusions, and wind tunnels are limited by the scale effect and test conditions and are difficult to be directly applied to the study of engineering-scale problems. Therefore, it is of great significance to effectively integrate simulation data with wind tunnel test data and improve the accuracy of aerodynamic noise prediction of engineering-scale protrusions.

[0003] At present, there are works that use neural networks to fuse simulation data with wind tunnel test data. However, the existing technologies have the following three problems when used to fuse simulation data with wind tunnel test data: 1) The complex data model established based on neural network methods is difficult to meet the physical constraints of fluid problems; 2) There are many geometric parameters and working condition description parameters of engineering problems, which easily lead to dimensional disasters. When there is no huge amount of data as a training set, the model confidence is low; 3) The interpretability of the model directly established based on neural network methods is poor. Therefore, it is of great significance to develop a method for predicting aerodynamic noise of engineering-scale protrusions that takes into account physical constraints, interpretability, and low problem dimensionality. Summary of the invention

[0004] The technical problem solved by the present invention is: to overcome the shortcomings of the prior art and propose a method for predicting the aerodynamic noise of engineering-scale protrusions based on data fusion. By fusing the wind tunnel test data of the scaled model and the simulation data of the scaled model and the engineering-scale model, the aerodynamic noise of the engineering-scale protrusions can be accurately predicted under the premise that it is difficult to obtain the engineering-scale test data in the design stage.

[0005] The technical solution of the present invention is:

[0006] A method for predicting aerodynamic noise of protrusions at an engineering scale based on data fusion, comprising:

[0007] S1: Within the use range of the wind tunnel equipment, based on the protrusion model to be predicted, design several protrusion models with different scale sizes;

[0008] S2: For each scaled-down protrusion model, a corresponding wind tunnel test model is manufactured and aerodynamic noise tests are carried out. The pressure signals under each test condition are collected by sensors and converted into the power spectrum density of noise to obtain a test data set containing the power spectrum density of noise.

[0009] Construct a protrusion simulation model, simulate and obtain the flow field of protrusions of various scaled sizes, collect statistics on the pressure data at the same position as the sensor and convert the power spectrum density of the noise to obtain a simulation data set including simulation conditions and noise power spectrum density;

[0010] S3: extracting features from the data of the test data set and the data of the simulation data set respectively, wherein the feature data include the total sound pressure level of the noise power spectrum, the power spectrum density of the natural frequencies of the first two modes of the aircraft where the protrusion is located, and the power spectrum density of the natural frequencies of the first mode of the protrusion, to form a test feature data set and a simulation feature data set;

[0011] S4: performing active subspace mapping on the feature data of the simulation feature data set to form a mapping of the simulation condition space to the low-dimensional active subspace;

[0012] S5: construct a recurrent neural network, use the mapping relationship between the active subspace vector and the simulation condition as the mapping from the input layer of the recurrent neural network to the first layer of the intermediate layer, define the simulation data set as the input, and define the test data set as the output;

[0013] S6: training a recurrent neural network using the test data set and simulation data set obtained in step S2 to obtain a model for predicting the test power spectrum density data based on the simulation power spectrum density data;

[0014] S7: For the protrusion model to be predicted, simulate and obtain the noise power spectrum density data at the sensor position of the engineering-scale protrusion, input the noise power spectrum density data into the model for predicting the test power spectrum density data based on the simulated power spectrum density data, and obtain the flight test noise power spectrum density data of the sensor of the engineering-scale protrusion.

[0015] Furthermore, in step S4, active subspace mapping is performed on the feature data of the simulation feature data set, and the specific method is as follows:

[0016] Construct the partial derivative matrix of the characteristic data variables to the simulation condition variables:

[0017]

[0018] Where m is the total number of simulation states considering different model scales and working conditions; φ is the characteristic data; are simulation conditions, including model scale L, Mach number Ma, pressure p and temperature T;

[0019] Perform singular value decomposition on the partial derivative matrix to obtain the eigenvectors d1, d2, …, d p Span the active subspace to form a mapping from the high-dimensional simulation condition space to the low-dimensional active subspace. The specific formula is:

[0020]

[0021] in, is the simulation condition of the active subspace after mapping.

[0022] Furthermore, the partial derivative matrix is ​​decomposed into singular values:

[0023]

[0024] Where A is the skew variance matrix, Λ is the singular value matrix, and D is the right characteristic matrix;

[0025]

[0026] Λ1 and Λ2 are block diagonal matrices, and the specific formula is:

[0027] Λ1=diag(λ1,λ2,…,λ p )

[0028] Λ2=diag(λ p+1 ,…,λ m )

[0029] The basis of the block is λ p >>λ p+1 , λ is a singular value, and the expression of D1 is:

[0030] D1=[d1,d2,…,d p ]

[0031] That is, we get the eigenvectors d1, d2, …, d p Span the active subspace.

[0032] Furthermore, the test data set and simulation data set obtained in step S2 are randomly divided into a training set and a validation set; based on the data of the training set, a recurrent neural network is trained to obtain a model for predicting the test power spectral density data based on the simulation power spectral density data; the model is verified based on the data of the validation set, and if the maximum deviation between the feature data output by the model and the corresponding test feature data obtained in step S3 exceeds a set threshold, the model is continuously optimized and the optimized model is verified again.

[0033] Furthermore, the mapping relationship between the active subspace vector and the simulation condition is used as the mapping from the neural network input layer to the first layer of the intermediate layer, and the mapping only acts on the simulation condition of the network input.

[0034] Furthermore, the set threshold is 10% to 20%.

[0035] Further, for a simulation data set including simulation conditions and noise power spectrum density, the simulation conditions include model scale, Mach number, pressure and temperature.

[0036] Furthermore, a protrusion simulation model is constructed based on the large eddy simulation method, and the flow field of protrusions of various scaled sizes is simulated and obtained using the computational fluid dynamics method.

[0037] A computer-readable storage medium stores a computer program, and the computer program implements the steps of the above method when executed by a processor.

[0038] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0039] The advantages of the present invention compared with the prior art are:

[0040] (1) The present invention solves the problem of the difficulty in accurately predicting the aerodynamic noise of engineering-scale protrusions caused by scale effect and simulation accuracy through a data fusion method based on a recurrent neural network. Compared with the traditional wind tunnel test research method that does not consider the scale effect, the present invention can more accurately predict the aerodynamic noise of engineering-scale protrusions, which is beneficial to improving the performance of the aircraft.

[0041] (2) The present invention proposes a recurrent neural network with active subspace mapping. Aiming at the prediction problem of heat flux density of engineering-scale protrusions, simulations of different model scales and different working conditions are carried out to form a data set containing simulation conditions and heat flux density. Feature extraction is performed on the data set to form a feature data set. Active subspace mapping is performed on the variables of the feature data set. The obtained low-dimensional mapping is used as the mapping of the input layer of the recurrent neural network to the first intermediate layer, thereby improving the physical constraints and interpretability of the neural network and improving the model confidence. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0043] Figure 1 This is a flow chart of a method for predicting aerodynamic noise of protrusions at an engineering scale based on data fusion according to an embodiment of the present invention;

[0044] Figure 2 A flow chart of a method for obtaining test data and simulation data of each scaled model according to an embodiment of the present invention;

[0045] Figure 3 A flow chart of a model method for predicting experimental power spectral density data based on simulated power spectral density data is established for an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0047] The present invention proposes a method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion, such as Figure 1 As shown, the specific steps include:

[0048] 1) The length of the protrusion is L0. Within the use range of the wind tunnel equipment, the design number of protrusion models arranged from small to large is n, and the scaled sizes {L1, L2, …, L n};

[0049] 2) For each scaled size, a wind tunnel test model is manufactured and aerodynamic noise tests are carried out. The pressure signals under each test condition are collected by sensors and converted into the power spectrum density of noise to obtain the test data set {M1,M2,…,M n},like Figure 2 As shown;

[0050] 3) Based on the large eddy simulation method, a protrusion simulation model is constructed. The flow field of protrusions of various scaled sizes is simulated and obtained using the computational fluid dynamics method. The pressure data at the sensor position is counted and the power spectrum density of the noise is converted to obtain a simulation data set {S1, S2, …, S n};

[0051] 4) Extract features from the wind tunnel test data. The feature data include the total sound pressure level of the noise power spectrum, the power spectrum density of the first two natural frequencies of the aircraft where the protrusion is located, and the power spectrum density of the first natural frequency of the protrusion mode, forming a wind tunnel test feature data set {CM1, CM2, …, CM n};

[0052] 5) Feature extraction is performed on the simulation data. The feature data also includes the total sound pressure level of the noise power spectrum, the power spectrum density of the natural frequencies of the first two modes of the aircraft where the protrusion is located, and the power spectrum density of the natural frequency of the first mode of the protrusion, forming a simulation feature data set {CS1, CS2, …, CS n};

[0053] 6) Perform active subspace mapping on the feature data of the simulation feature data set. The specific method is: construct a partial derivative matrix of the feature data variable to the simulation condition variable. The specific formula is as follows:

[0054]

[0055] Where m is the total number of simulation states considering different model scales and working conditions, Φ is the characteristic data, is the simulation condition, including model scale L, Mach number Ma, pressure p and temperature T. The specific formula is as follows:

[0056]

[0057] Perform singular value decomposition on the partial derivative matrix. The specific formula is:

[0058]

[0059] Among them, A is the skew variance matrix, Λ is the singular value matrix, and D is the right eigenvalue matrix. The singular value matrix and the right eigenvalue matrix are divided into blocks according to the size of the singular values. The specific formula is:

[0060]

[0061] Where Λ1 and Λ2 are block diagonal matrices, the specific formula is:

[0062] Λ1=diag(λ1,λ2,…,λ p )

[0063] Λ2=diag(λ p+1 ,…,λ m )

[0064] The basis of the block is λ p >>λ p+1 , λ is a singular value, and the expression of D1 is:

[0065] D1=[d1,d2,…,d p ]

[0066] Eigenvectors d1,d2,…,d p Span the active subspace to form a mapping from the high-dimensional simulation condition space to the low-dimensional active subspace. The specific formula is:

[0067]

[0068] in, is the simulation condition of the active subspace after mapping.

[0069] 7) Construct a recurrent neural network, and use the mapping relationship between the active subspace vector and the simulation condition as the mapping from the neural network input layer to the first layer of the intermediate layer (the mapping only acts on the simulation condition of the model input). n} as the input of the model, and the experimental data set {M1,M2,…,M n} as the output of the model, such as Figure 3 As shown in the figure, according to the scaled size and working conditions, the input and output are randomly divided into a neural network training set, a validation set and a test set. The neural network is trained, and a model based on the simulated power spectrum density data to predict the test power spectrum density data is obtained. The output of the model is tested by the validation set and the test set. The power spectrum density feature data of the model output is obtained by feature extraction. If the feature data of the model output is consistent with the feature data of the validation set {CM1, CM2, …, CM n If the maximum deviation of} exceeds 15%, continue to optimize the model;

[0070] 8) Simulate and obtain the flow field of the engineering-scale protrusion, and obtain the noise power spectrum density data S0 at the sensor position;

[0071] 9) By using a model that predicts test power spectrum density data based on simulation power spectrum density data, the noise power spectrum density data S0 at the sensor position of the engineering scale protrusion is used as input to obtain the flight test noise power spectrum density data M0 of the sensor of the engineering scale protrusion.

[0072] The present invention solves the problem of accurate prediction of aerodynamic noise of engineering-scale protrusions caused by scale effect and simulation accuracy through a data fusion method based on a recurrent neural network. Compared with the traditional wind tunnel test research method that does not consider the scale effect, the present invention can more accurately predict the aerodynamic noise of engineering-scale protrusions, which is beneficial to improving aircraft performance.

[0073] The above-described embodiments are only preferred specific implementations of the present invention. Common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion, characterized in that: include: S1: Within the use range of the wind tunnel equipment, based on the protrusion model to be predicted, design several protrusion models with different scale sizes; S2: For each scaled-down protrusion model, a corresponding wind tunnel test model is manufactured and aerodynamic noise tests are carried out. The pressure signals under each test condition are collected by sensors and converted into the power spectrum density of noise to obtain a test data set containing the power spectrum density of noise. Construct a protrusion simulation model, simulate and obtain the flow field of protrusions of various scaled sizes, collect statistics on the pressure data at the same position as the sensor and convert the power spectrum density of the noise to obtain a simulation data set including simulation conditions and noise power spectrum density; S3: extracting features from the data of the test data set and the data of the simulation data set respectively, wherein the feature data include the total sound pressure level of the noise power spectrum, the power spectrum density of the natural frequencies of the first two modes of the aircraft where the protrusion is located, and the power spectrum density of the natural frequencies of the first mode of the protrusion, to form a test feature data set and a simulation feature data set; S4: performing active subspace mapping on the feature data of the simulation feature data set to form a mapping of the simulation condition space to the low-dimensional active subspace; S5: construct a recurrent neural network, use the mapping relationship between the active subspace vector and the simulation condition as the mapping from the input layer of the recurrent neural network to the first layer of the intermediate layer, define the simulation data set as the input, and define the test data set as the output; S6: training a recurrent neural network using the test data set and simulation data set obtained in step S2 to obtain a model for predicting the test power spectrum density data based on the simulation power spectrum density data; S7: For the protrusion model to be predicted, simulate and obtain the noise power spectrum density data at the sensor position of the engineering-scale protrusion, input the noise power spectrum density data into the model for predicting the test power spectrum density data based on the simulated power spectrum density data, and obtain the flight test noise power spectrum density data of the sensor of the engineering-scale protrusion.

2. The method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion according to claim 1, characterized in that: In step S4, active subspace mapping is performed on the feature data of the simulation feature data set, and the specific method is as follows: Construct the partial derivative matrix of the characteristic data variables to the simulation condition variables: Where m is the total number of simulation states considering different model scales and working conditions; φ is the characteristic data; are simulation conditions, including model scale L, Mach number Ma, pressure p and temperature T; Perform singular value decomposition on the partial derivative matrix to obtain the eigenvectors d1, d2, …, d p Span the active subspace to form a mapping from the high-dimensional simulation condition space to the low-dimensional active subspace. The specific formula is: Among them, ζ is the simulation condition of the active subspace after mapping.

3. The method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion according to claim 2, characterized in that: Perform singular value decomposition on the partial derivative matrix: Where A is the skew variance matrix, Λ is the singular value matrix, and D is the right characteristic matrix; D=[D1,D2] Λ1 and Λ2 are block diagonal matrices, and the specific formula is: Λ1=diag(λ1,λ2,…,λ p ) Λ2=diag(λ p+1 ,…,l m ) The basis of the block is λ p >>λ p+1 , λ is a singular value, and the expression of D1 is: D1=[d1,d2,…,d p ] That is, we get the eigenvectors d1, d2, …, d p Span the active subspace.

4. The method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion according to claim 1, characterized in that: The test data set and simulation data set obtained in step S2 are randomly divided into a training set and a validation set; based on the data of the training set, a recurrent neural network is trained to obtain a model that predicts the test power spectral density data based on the simulation power spectral density data; the model is verified based on the data of the validation set, and if the maximum deviation between the feature data output by the model and the corresponding test feature data obtained in step S3 exceeds a set threshold, the model is continuously optimized and the optimized model is verified again.

5. The method for predicting aerodynamic noise of protrusions at an engineering scale based on data fusion according to any one of claims 1 to 4, characterized in that: The mapping relationship between the active subspace vector and the simulation condition is used as the mapping from the neural network input layer to the first layer of the intermediate layer. The mapping only acts on the simulation condition of the network input.

6. The method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion according to claim 4, characterized in that: The set threshold is 10% to 20%.

7. The method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion according to claim 1, characterized in that: For a simulation data set including simulation conditions and noise power spectrum density, the simulation conditions include model scale, Mach number, pressure and temperature.

8. The method for predicting aerodynamic noise of engineering-scale protrusions based on data fusion according to claim 1, characterized in that: A protrusion simulation model is constructed based on the large eddy simulation method, and the flow field of protrusions of various scaled sizes is simulated and obtained using the computational fluid dynamics method.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images